Vals AI Updates

AI cheating is on the rise

Introducing Terminal-Bench Science: research workflows written by working scientists

DeepSeek's V4.1 Flash evaluated across our benchmark suite

Claude Fable 5.1 ranks #1 on the RSI Index

Introducing Tax Agent Bench: research-grade US tax questions for agents

Vals Environmental Impacts Report

Anthropic's Claude Fable 5.1 evaluated across our benchmark suite

Claude Fable 5.1 solves a 370-year-old cipher

Meta's Muse Voice Transcribe evaluated on VoiceCodeBench

Introducing VoiceCodeBench: can speech-to-text preserve the values a workflow depends on?